2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_trans_priv.h"
41 #include "xfs_error.h"
45 STATIC xfs_buf_t *xfs_trans_buf_item_match(xfs_trans_t *, xfs_buftarg_t *,
47 STATIC xfs_buf_t *xfs_trans_buf_item_match_all(xfs_trans_t *, xfs_buftarg_t *,
52 * Get and lock the buffer for the caller if it is not already
53 * locked within the given transaction. If it is already locked
54 * within the transaction, just increment its lock recursion count
55 * and return a pointer to it.
57 * Use the fast path function xfs_trans_buf_item_match() or the buffer
58 * cache routine incore_match() to find the buffer
59 * if it is already owned by this transaction.
61 * If we don't already own the buffer, use get_buf() to get it.
62 * If it doesn't yet have an associated xfs_buf_log_item structure,
63 * then allocate one and add the item to this transaction.
65 * If the transaction pointer is NULL, make this just a normal
69 xfs_trans_get_buf(xfs_trans_t *tp,
70 xfs_buftarg_t *target_dev,
76 xfs_buf_log_item_t *bip;
79 flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
82 * Default to a normal get_buf() call if the tp is NULL.
85 bp = xfs_buf_get_flags(target_dev, blkno, len,
91 * If we find the buffer in the cache with this transaction
92 * pointer in its b_fsprivate2 field, then we know we already
93 * have it locked. In this case we just increment the lock
94 * recursion count and return the buffer to the caller.
96 if (tp->t_items.lic_next == NULL) {
97 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
99 bp = xfs_trans_buf_item_match_all(tp, target_dev, blkno, len);
102 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
103 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
104 xfs_buftrace("TRANS GET RECUR SHUT", bp);
105 XFS_BUF_SUPER_STALE(bp);
108 * If the buffer is stale then it was binval'ed
109 * since last read. This doesn't matter since the
110 * caller isn't allowed to use the data anyway.
112 else if (XFS_BUF_ISSTALE(bp)) {
113 xfs_buftrace("TRANS GET RECUR STALE", bp);
114 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
116 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
117 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
119 ASSERT(atomic_read(&bip->bli_refcount) > 0);
121 xfs_buftrace("TRANS GET RECUR", bp);
122 xfs_buf_item_trace("GET RECUR", bip);
127 * We always specify the BUF_BUSY flag within a transaction so
128 * that get_buf does not try to push out a delayed write buffer
129 * which might cause another transaction to take place (if the
130 * buffer was delayed alloc). Such recursive transactions can
131 * easily deadlock with our current transaction as well as cause
132 * us to run out of stack space.
134 bp = xfs_buf_get_flags(target_dev, blkno, len, flags | BUF_BUSY);
139 ASSERT(!XFS_BUF_GETERROR(bp));
142 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
143 * it doesn't have one yet, then allocate one and initialize it.
144 * The checks to see if one is there are in xfs_buf_item_init().
146 xfs_buf_item_init(bp, tp->t_mountp);
149 * Set the recursion count for the buffer within this transaction
152 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
153 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
154 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
155 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
159 * Take a reference for this transaction on the buf item.
161 atomic_inc(&bip->bli_refcount);
164 * Get a log_item_desc to point at the new item.
166 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
169 * Initialize b_fsprivate2 so we can find it with incore_match()
172 XFS_BUF_SET_FSPRIVATE2(bp, tp);
174 xfs_buftrace("TRANS GET", bp);
175 xfs_buf_item_trace("GET", bip);
180 * Get and lock the superblock buffer of this file system for the
183 * We don't need to use incore_match() here, because the superblock
184 * buffer is a private buffer which we keep a pointer to in the
188 xfs_trans_getsb(xfs_trans_t *tp,
189 struct xfs_mount *mp,
193 xfs_buf_log_item_t *bip;
196 * Default to just trying to lock the superblock buffer
200 return (xfs_getsb(mp, flags));
204 * If the superblock buffer already has this transaction
205 * pointer in its b_fsprivate2 field, then we know we already
206 * have it locked. In this case we just increment the lock
207 * recursion count and return the buffer to the caller.
210 if (XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp) {
211 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
213 ASSERT(atomic_read(&bip->bli_refcount) > 0);
215 xfs_buf_item_trace("GETSB RECUR", bip);
219 bp = xfs_getsb(mp, flags);
225 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
226 * it doesn't have one yet, then allocate one and initialize it.
227 * The checks to see if one is there are in xfs_buf_item_init().
229 xfs_buf_item_init(bp, mp);
232 * Set the recursion count for the buffer within this transaction
235 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
236 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
237 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
238 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
242 * Take a reference for this transaction on the buf item.
244 atomic_inc(&bip->bli_refcount);
247 * Get a log_item_desc to point at the new item.
249 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
252 * Initialize b_fsprivate2 so we can find it with incore_match()
255 XFS_BUF_SET_FSPRIVATE2(bp, tp);
257 xfs_buf_item_trace("GETSB", bip);
262 xfs_buftarg_t *xfs_error_target;
265 int xfs_error_mod = 33;
269 * Get and lock the buffer for the caller if it is not already
270 * locked within the given transaction. If it has not yet been
271 * read in, read it from disk. If it is already locked
272 * within the transaction and already read in, just increment its
273 * lock recursion count and return a pointer to it.
275 * Use the fast path function xfs_trans_buf_item_match() or the buffer
276 * cache routine incore_match() to find the buffer
277 * if it is already owned by this transaction.
279 * If we don't already own the buffer, use read_buf() to get it.
280 * If it doesn't yet have an associated xfs_buf_log_item structure,
281 * then allocate one and add the item to this transaction.
283 * If the transaction pointer is NULL, make this just a normal
290 xfs_buftarg_t *target,
297 xfs_buf_log_item_t *bip;
301 flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
304 * Default to a normal get_buf() call if the tp is NULL.
307 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
309 return XFS_ERROR(ENOMEM);
311 if ((bp != NULL) && (XFS_BUF_GETERROR(bp) != 0)) {
312 xfs_ioerror_alert("xfs_trans_read_buf", mp,
314 error = XFS_BUF_GETERROR(bp);
319 if (xfs_do_error && (bp != NULL)) {
320 if (xfs_error_target == target) {
321 if (((xfs_req_num++) % xfs_error_mod) == 0) {
323 printk("Returning error!\n");
324 return XFS_ERROR(EIO);
329 if (XFS_FORCED_SHUTDOWN(mp))
336 * If we find the buffer in the cache with this transaction
337 * pointer in its b_fsprivate2 field, then we know we already
338 * have it locked. If it is already read in we just increment
339 * the lock recursion count and return the buffer to the caller.
340 * If the buffer is not yet read in, then we read it in, increment
341 * the lock recursion count, and return it to the caller.
343 if (tp->t_items.lic_next == NULL) {
344 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
346 bp = xfs_trans_buf_item_match_all(tp, target, blkno, len);
349 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
350 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
351 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
352 ASSERT((XFS_BUF_ISERROR(bp)) == 0);
353 if (!(XFS_BUF_ISDONE(bp))) {
354 xfs_buftrace("READ_BUF_INCORE !DONE", bp);
355 ASSERT(!XFS_BUF_ISASYNC(bp));
357 xfsbdstrat(tp->t_mountp, bp);
359 if (XFS_BUF_GETERROR(bp) != 0) {
360 xfs_ioerror_alert("xfs_trans_read_buf", mp,
362 error = XFS_BUF_GETERROR(bp);
365 * We can gracefully recover from most
366 * read errors. Ones we can't are those
367 * that happen after the transaction's
370 if (tp->t_flags & XFS_TRANS_DIRTY)
371 xfs_force_shutdown(tp->t_mountp,
372 XFS_METADATA_IO_ERROR);
377 * We never locked this buf ourselves, so we shouldn't
378 * brelse it either. Just get out.
380 if (XFS_FORCED_SHUTDOWN(mp)) {
381 xfs_buftrace("READ_BUF_INCORE XFSSHUTDN", bp);
383 return XFS_ERROR(EIO);
387 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
390 ASSERT(atomic_read(&bip->bli_refcount) > 0);
391 xfs_buf_item_trace("READ RECUR", bip);
397 * We always specify the BUF_BUSY flag within a transaction so
398 * that get_buf does not try to push out a delayed write buffer
399 * which might cause another transaction to take place (if the
400 * buffer was delayed alloc). Such recursive transactions can
401 * easily deadlock with our current transaction as well as cause
402 * us to run out of stack space.
404 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
409 if (XFS_BUF_GETERROR(bp) != 0) {
410 XFS_BUF_SUPER_STALE(bp);
411 xfs_buftrace("READ ERROR", bp);
412 error = XFS_BUF_GETERROR(bp);
414 xfs_ioerror_alert("xfs_trans_read_buf", mp,
416 if (tp->t_flags & XFS_TRANS_DIRTY)
417 xfs_force_shutdown(tp->t_mountp, XFS_METADATA_IO_ERROR);
422 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
423 if (xfs_error_target == target) {
424 if (((xfs_req_num++) % xfs_error_mod) == 0) {
425 xfs_force_shutdown(tp->t_mountp,
426 XFS_METADATA_IO_ERROR);
428 printk("Returning error in trans!\n");
429 return XFS_ERROR(EIO);
434 if (XFS_FORCED_SHUTDOWN(mp))
438 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
439 * it doesn't have one yet, then allocate one and initialize it.
440 * The checks to see if one is there are in xfs_buf_item_init().
442 xfs_buf_item_init(bp, tp->t_mountp);
445 * Set the recursion count for the buffer within this transaction
448 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
449 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
450 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
451 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
455 * Take a reference for this transaction on the buf item.
457 atomic_inc(&bip->bli_refcount);
460 * Get a log_item_desc to point at the new item.
462 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
465 * Initialize b_fsprivate2 so we can find it with incore_match()
468 XFS_BUF_SET_FSPRIVATE2(bp, tp);
470 xfs_buftrace("TRANS READ", bp);
471 xfs_buf_item_trace("READ", bip);
477 * the theory here is that buffer is good but we're
478 * bailing out because the filesystem is being forcibly
479 * shut down. So we should leave the b_flags alone since
480 * the buffer's not staled and just get out.
483 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
484 cmn_err(CE_NOTE, "about to pop assert, bp == 0x%p", bp);
486 ASSERT((XFS_BUF_BFLAGS(bp) & (XFS_B_STALE|XFS_B_DELWRI)) !=
487 (XFS_B_STALE|XFS_B_DELWRI));
489 xfs_buftrace("READ_BUF XFSSHUTDN", bp);
492 return XFS_ERROR(EIO);
497 * Release the buffer bp which was previously acquired with one of the
498 * xfs_trans_... buffer allocation routines if the buffer has not
499 * been modified within this transaction. If the buffer is modified
500 * within this transaction, do decrement the recursion count but do
501 * not release the buffer even if the count goes to 0. If the buffer is not
502 * modified within the transaction, decrement the recursion count and
503 * release the buffer if the recursion count goes to 0.
505 * If the buffer is to be released and it was not modified before
506 * this transaction began, then free the buf_log_item associated with it.
508 * If the transaction pointer is NULL, make this just a normal
512 xfs_trans_brelse(xfs_trans_t *tp,
515 xfs_buf_log_item_t *bip;
517 xfs_log_item_desc_t *lidp;
520 * Default to a normal brelse() call if the tp is NULL.
523 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
525 * If there's a buf log item attached to the buffer,
526 * then let the AIL know that the buffer is being
529 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
530 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
531 if (lip->li_type == XFS_LI_BUF) {
532 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
533 xfs_trans_unlocked_item(
534 bip->bli_item.li_mountp,
542 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
543 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
544 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
545 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
546 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
547 ASSERT(atomic_read(&bip->bli_refcount) > 0);
550 * Find the item descriptor pointing to this buffer's
551 * log item. It must be there.
553 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
554 ASSERT(lidp != NULL);
557 * If the release is just for a recursive lock,
558 * then decrement the count and return.
560 if (bip->bli_recur > 0) {
562 xfs_buf_item_trace("RELSE RECUR", bip);
567 * If the buffer is dirty within this transaction, we can't
568 * release it until we commit.
570 if (lidp->lid_flags & XFS_LID_DIRTY) {
571 xfs_buf_item_trace("RELSE DIRTY", bip);
576 * If the buffer has been invalidated, then we can't release
577 * it until the transaction commits to disk unless it is re-dirtied
578 * as part of this transaction. This prevents us from pulling
579 * the item from the AIL before we should.
581 if (bip->bli_flags & XFS_BLI_STALE) {
582 xfs_buf_item_trace("RELSE STALE", bip);
586 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
587 xfs_buf_item_trace("RELSE", bip);
590 * Free up the log item descriptor tracking the released item.
592 xfs_trans_free_item(tp, lidp);
595 * Clear the hold flag in the buf log item if it is set.
596 * We wouldn't want the next user of the buffer to
599 if (bip->bli_flags & XFS_BLI_HOLD) {
600 bip->bli_flags &= ~XFS_BLI_HOLD;
604 * Drop our reference to the buf log item.
606 atomic_dec(&bip->bli_refcount);
609 * If the buf item is not tracking data in the log, then
610 * we must free it before releasing the buffer back to the
611 * free pool. Before releasing the buffer to the free pool,
612 * clear the transaction pointer in b_fsprivate2 to dissolve
613 * its relation to this transaction.
615 if (!xfs_buf_item_dirty(bip)) {
617 ASSERT(bp->b_pincount == 0);
619 ASSERT(atomic_read(&bip->bli_refcount) == 0);
620 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
621 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
622 xfs_buf_item_relse(bp);
625 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
628 * If we've still got a buf log item on the buffer, then
629 * tell the AIL that the buffer is being unlocked.
632 xfs_trans_unlocked_item(bip->bli_item.li_mountp,
633 (xfs_log_item_t*)bip);
641 * Add the locked buffer to the transaction.
642 * The buffer must be locked, and it cannot be associated with any
645 * If the buffer does not yet have a buf log item associated with it,
646 * then allocate one for it. Then add the buf item to the transaction.
649 xfs_trans_bjoin(xfs_trans_t *tp,
652 xfs_buf_log_item_t *bip;
654 ASSERT(XFS_BUF_ISBUSY(bp));
655 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
658 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
659 * it doesn't have one yet, then allocate one and initialize it.
660 * The checks to see if one is there are in xfs_buf_item_init().
662 xfs_buf_item_init(bp, tp->t_mountp);
663 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
664 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
665 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
666 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
669 * Take a reference for this transaction on the buf item.
671 atomic_inc(&bip->bli_refcount);
674 * Get a log_item_desc to point at the new item.
676 (void) xfs_trans_add_item(tp, (xfs_log_item_t *)bip);
679 * Initialize b_fsprivate2 so we can find it with incore_match()
680 * in xfs_trans_get_buf() and friends above.
682 XFS_BUF_SET_FSPRIVATE2(bp, tp);
684 xfs_buf_item_trace("BJOIN", bip);
688 * Mark the buffer as not needing to be unlocked when the buf item's
689 * IOP_UNLOCK() routine is called. The buffer must already be locked
690 * and associated with the given transaction.
694 xfs_trans_bhold(xfs_trans_t *tp,
697 xfs_buf_log_item_t *bip;
699 ASSERT(XFS_BUF_ISBUSY(bp));
700 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
701 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
703 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
704 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
705 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
706 ASSERT(atomic_read(&bip->bli_refcount) > 0);
707 bip->bli_flags |= XFS_BLI_HOLD;
708 xfs_buf_item_trace("BHOLD", bip);
712 * Cancel the previous buffer hold request made on this buffer
713 * for this transaction.
716 xfs_trans_bhold_release(xfs_trans_t *tp,
719 xfs_buf_log_item_t *bip;
721 ASSERT(XFS_BUF_ISBUSY(bp));
722 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
723 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
725 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
726 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
727 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
728 ASSERT(atomic_read(&bip->bli_refcount) > 0);
729 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
730 bip->bli_flags &= ~XFS_BLI_HOLD;
731 xfs_buf_item_trace("BHOLD RELEASE", bip);
735 * This is called to mark bytes first through last inclusive of the given
736 * buffer as needing to be logged when the transaction is committed.
737 * The buffer must already be associated with the given transaction.
739 * First and last are numbers relative to the beginning of this buffer,
740 * so the first byte in the buffer is numbered 0 regardless of the
744 xfs_trans_log_buf(xfs_trans_t *tp,
749 xfs_buf_log_item_t *bip;
750 xfs_log_item_desc_t *lidp;
752 ASSERT(XFS_BUF_ISBUSY(bp));
753 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
754 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
755 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
756 ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) ||
757 (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks));
760 * Mark the buffer as needing to be written out eventually,
761 * and set its iodone function to remove the buffer's buf log
762 * item from the AIL and free it when the buffer is flushed
763 * to disk. See xfs_buf_attach_iodone() for more details
764 * on li_cb and xfs_buf_iodone_callbacks().
765 * If we end up aborting this transaction, we trap this buffer
766 * inside the b_bdstrat callback so that this won't get written to
769 XFS_BUF_DELAYWRITE(bp);
772 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
773 ASSERT(atomic_read(&bip->bli_refcount) > 0);
774 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
775 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))xfs_buf_iodone;
778 * If we invalidated the buffer within this transaction, then
779 * cancel the invalidation now that we're dirtying the buffer
780 * again. There are no races with the code in xfs_buf_item_unpin(),
781 * because we have a reference to the buffer this entire time.
783 if (bip->bli_flags & XFS_BLI_STALE) {
784 xfs_buf_item_trace("BLOG UNSTALE", bip);
785 bip->bli_flags &= ~XFS_BLI_STALE;
786 ASSERT(XFS_BUF_ISSTALE(bp));
788 bip->bli_format.blf_flags &= ~XFS_BLI_CANCEL;
791 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
792 ASSERT(lidp != NULL);
794 tp->t_flags |= XFS_TRANS_DIRTY;
795 lidp->lid_flags |= XFS_LID_DIRTY;
796 lidp->lid_flags &= ~XFS_LID_BUF_STALE;
797 bip->bli_flags |= XFS_BLI_LOGGED;
798 xfs_buf_item_log(bip, first, last);
799 xfs_buf_item_trace("BLOG", bip);
804 * This called to invalidate a buffer that is being used within
805 * a transaction. Typically this is because the blocks in the
806 * buffer are being freed, so we need to prevent it from being
807 * written out when we're done. Allowing it to be written again
808 * might overwrite data in the free blocks if they are reallocated
811 * We prevent the buffer from being written out by clearing the
812 * B_DELWRI flag. We can't always
813 * get rid of the buf log item at this point, though, because
814 * the buffer may still be pinned by another transaction. If that
815 * is the case, then we'll wait until the buffer is committed to
816 * disk for the last time (we can tell by the ref count) and
817 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
818 * will keep the buffer locked so that the buffer and buf log item
826 xfs_log_item_desc_t *lidp;
827 xfs_buf_log_item_t *bip;
829 ASSERT(XFS_BUF_ISBUSY(bp));
830 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
831 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
833 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
834 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
835 ASSERT(lidp != NULL);
836 ASSERT(atomic_read(&bip->bli_refcount) > 0);
838 if (bip->bli_flags & XFS_BLI_STALE) {
840 * If the buffer is already invalidated, then
843 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
844 ASSERT(XFS_BUF_ISSTALE(bp));
845 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
846 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_INODE_BUF));
847 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
848 ASSERT(lidp->lid_flags & XFS_LID_DIRTY);
849 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
850 xfs_buftrace("XFS_BINVAL RECUR", bp);
851 xfs_buf_item_trace("BINVAL RECUR", bip);
856 * Clear the dirty bit in the buffer and set the STALE flag
857 * in the buf log item. The STALE flag will be used in
858 * xfs_buf_item_unpin() to determine if it should clean up
859 * when the last reference to the buf item is given up.
860 * We set the XFS_BLI_CANCEL flag in the buf log format structure
861 * and log the buf item. This will be used at recovery time
862 * to determine that copies of the buffer in the log before
863 * this should not be replayed.
864 * We mark the item descriptor and the transaction dirty so
865 * that we'll hold the buffer until after the commit.
867 * Since we're invalidating the buffer, we also clear the state
868 * about which parts of the buffer have been logged. We also
869 * clear the flag indicating that this is an inode buffer since
870 * the data in the buffer will no longer be valid.
872 * We set the stale bit in the buffer as well since we're getting
875 XFS_BUF_UNDELAYWRITE(bp);
877 bip->bli_flags |= XFS_BLI_STALE;
878 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_DIRTY);
879 bip->bli_format.blf_flags &= ~XFS_BLI_INODE_BUF;
880 bip->bli_format.blf_flags |= XFS_BLI_CANCEL;
881 memset((char *)(bip->bli_format.blf_data_map), 0,
882 (bip->bli_format.blf_map_size * sizeof(uint)));
883 lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE;
884 tp->t_flags |= XFS_TRANS_DIRTY;
885 xfs_buftrace("XFS_BINVAL", bp);
886 xfs_buf_item_trace("BINVAL", bip);
890 * This call is used to indicate that the buffer contains on-disk
891 * inodes which must be handled specially during recovery. They
892 * require special handling because only the di_next_unlinked from
893 * the inodes in the buffer should be recovered. The rest of the
894 * data in the buffer is logged via the inodes themselves.
896 * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log
897 * format structure so that we'll know what to do at recovery time.
905 xfs_buf_log_item_t *bip;
907 ASSERT(XFS_BUF_ISBUSY(bp));
908 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
909 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
911 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
912 ASSERT(atomic_read(&bip->bli_refcount) > 0);
914 bip->bli_format.blf_flags |= XFS_BLI_INODE_BUF;
918 * This call is used to indicate that the buffer is going to
919 * be staled and was an inode buffer. This means it gets
920 * special processing during unpin - where any inodes
921 * associated with the buffer should be removed from ail.
922 * There is also special processing during recovery,
923 * any replay of the inodes in the buffer needs to be
924 * prevented as the buffer may have been reused.
927 xfs_trans_stale_inode_buf(
931 xfs_buf_log_item_t *bip;
933 ASSERT(XFS_BUF_ISBUSY(bp));
934 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
935 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
937 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
938 ASSERT(atomic_read(&bip->bli_refcount) > 0);
940 bip->bli_flags |= XFS_BLI_STALE_INODE;
941 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))
948 * Mark the buffer as being one which contains newly allocated
949 * inodes. We need to make sure that even if this buffer is
950 * relogged as an 'inode buf' we still recover all of the inode
951 * images in the face of a crash. This works in coordination with
952 * xfs_buf_item_committed() to ensure that the buffer remains in the
953 * AIL at its original location even after it has been relogged.
957 xfs_trans_inode_alloc_buf(
961 xfs_buf_log_item_t *bip;
963 ASSERT(XFS_BUF_ISBUSY(bp));
964 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
965 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
967 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
968 ASSERT(atomic_read(&bip->bli_refcount) > 0);
970 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
975 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
976 * dquots. However, unlike in inode buffer recovery, dquot buffers get
977 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
978 * The only thing that makes dquot buffers different from regular
979 * buffers is that we must not replay dquot bufs when recovering
980 * if a _corresponding_ quotaoff has happened. We also have to distinguish
981 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
982 * can be turned off independently.
991 xfs_buf_log_item_t *bip;
993 ASSERT(XFS_BUF_ISBUSY(bp));
994 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
995 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
996 ASSERT(type == XFS_BLI_UDQUOT_BUF ||
997 type == XFS_BLI_PDQUOT_BUF ||
998 type == XFS_BLI_GDQUOT_BUF);
1000 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
1001 ASSERT(atomic_read(&bip->bli_refcount) > 0);
1003 bip->bli_format.blf_flags |= type;
1007 * Check to see if a buffer matching the given parameters is already
1008 * a part of the given transaction. Only check the first, embedded
1009 * chunk, since we don't want to spend all day scanning large transactions.
1012 xfs_trans_buf_item_match(
1014 xfs_buftarg_t *target,
1018 xfs_log_item_chunk_t *licp;
1019 xfs_log_item_desc_t *lidp;
1020 xfs_buf_log_item_t *blip;
1026 licp = &tp->t_items;
1027 if (!XFS_LIC_ARE_ALL_FREE(licp)) {
1028 for (i = 0; i < licp->lic_unused; i++) {
1030 * Skip unoccupied slots.
1032 if (XFS_LIC_ISFREE(licp, i)) {
1036 lidp = XFS_LIC_SLOT(licp, i);
1037 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1038 if (blip->bli_item.li_type != XFS_LI_BUF) {
1043 if ((XFS_BUF_TARGET(bp) == target) &&
1044 (XFS_BUF_ADDR(bp) == blkno) &&
1045 (XFS_BUF_COUNT(bp) == len)) {
1047 * We found it. Break out and
1048 * return the pointer to the buffer.
1060 * Check to see if a buffer matching the given parameters is already
1061 * a part of the given transaction. Check all the chunks, we
1062 * want to be thorough.
1065 xfs_trans_buf_item_match_all(
1067 xfs_buftarg_t *target,
1071 xfs_log_item_chunk_t *licp;
1072 xfs_log_item_desc_t *lidp;
1073 xfs_buf_log_item_t *blip;
1079 for (licp = &tp->t_items; licp != NULL; licp = licp->lic_next) {
1080 if (XFS_LIC_ARE_ALL_FREE(licp)) {
1081 ASSERT(licp == &tp->t_items);
1082 ASSERT(licp->lic_next == NULL);
1085 for (i = 0; i < licp->lic_unused; i++) {
1087 * Skip unoccupied slots.
1089 if (XFS_LIC_ISFREE(licp, i)) {
1093 lidp = XFS_LIC_SLOT(licp, i);
1094 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1095 if (blip->bli_item.li_type != XFS_LI_BUF) {
1100 if ((XFS_BUF_TARGET(bp) == target) &&
1101 (XFS_BUF_ADDR(bp) == blkno) &&
1102 (XFS_BUF_COUNT(bp) == len)) {
1104 * We found it. Break out and
1105 * return the pointer to the buffer.